Flood Prediction In Malawi
Karim Mezhoud
2020-03-04
- 0.1 Delimitate the area of the study
- 1 Glimpse to Hydrology, Elevation, Landscover, and Population Maps
- 1.1 Glimpse on the Altitude (Elevation) and Type soil in the region using train dataset
- 1.2 Omit text and axis from ggplot
- 1.3 Preprocessing Train data
- 1.4 Set Water bodies, Permanent Wetland and Rivers areas with independent class even that have the same Altitude with other Soil Name
- 1.5 Visualize Pluviometry
- 1.6 Which Soil is more in flood?
- 1.7 Weight for Soil/Height/Flood
- 1.8 Categorize elevation to set slopes and define how water moves in the environment
- 1.9 Target distribution
- 2 Deal with unbalanced dataset
- 3 Random Forest modeling of Flood Zone Probability
- 4 Caret
- 5 Round Target to (0,1)
- 6 Animated Pluviometry of 2015 and 2019
- 7 Xgboost regression with Python
# Set python environment and version in RStudio ;-)
reticulate::use_python("/Users/Mezhoud/anaconda3/bin/python3", required = TRUE)
reticulate::py_config()## python: /Users/Mezhoud/anaconda3/bin/python3
## libpython: /Users/Mezhoud/anaconda3/lib/libpython3.7m.dylib
## pythonhome: /Users/Mezhoud/anaconda3:/Users/Mezhoud/anaconda3
## version: 3.7.5 (default, Oct 25 2019, 10:52:18) [Clang 4.0.1 (tags/RELEASE_401/final)]
## numpy: /Users/Mezhoud/anaconda3/lib/python3.7/site-packages/numpy
## numpy_version: 1.18.1
##
## NOTE: Python version was forced by use_python functionTrain <- fread("Train.csv")
Train <- Train %>%
rename(`Altitude (m)` = elevation) %>%
rename(Longitude = X) %>%
rename(Latitude = Y) %>%
separate(Square_ID, into = c("Square_ID", "v1", "v2", "v3", "v4"), sep = "-" , remove = FALSE) %>%
unite("other", v1, v2,v3, v4, sep = "-") %>%
select(Longitude,Latitude,`Altitude (m)`, LC_Type1_mode,Square_ID, other , everything())
#filter(row_number()==1 )
Train %>% head()## Longitude Latitude Altitude (m) LC_Type1_mode Square_ID
## 1: 34.26 -15.91 887.7642 9 4e3c3896
## 2: 34.26 -15.90 743.4039 9 4e3c3897
## 3: 34.26 -15.89 565.7283 9 4e3c3898
## 4: 34.26 -15.88 443.3928 10 4e3c3899
## 5: 34.26 -15.87 437.4434 10 4e3c389a
## 6: 34.26 -15.86 405.6317 10 4e3c389b
## other target_2015 precip 2014-11-16 - 2014-11-23
## 1: 14ce-11ea-bce5-f49634744a41 0 0
## 2: 14ce-11ea-bce5-f49634744a41 0 0
## 3: 14ce-11ea-bce5-f49634744a41 0 0
## 4: 14ce-11ea-bce5-f49634744a41 0 0
## 5: 14ce-11ea-bce5-f49634744a41 0 0
## 6: 14ce-11ea-bce5-f49634744a41 0 0
## precip 2014-11-23 - 2014-11-30 precip 2014-11-30 - 2014-12-07
## 1: 0 0
## 2: 0 0
## 3: 0 0
## 4: 0 0
## 5: 0 0
## 6: 0 0
## precip 2014-12-07 - 2014-12-14 precip 2014-12-14 - 2014-12-21
## 1: 14.84403 14.55282
## 2: 14.84403 14.55282
## 3: 14.84403 14.55282
## 4: 14.84403 14.55282
## 5: 14.84403 14.55282
## 6: 14.84403 14.55282
## precip 2014-12-21 - 2014-12-28 precip 2014-12-28 - 2015-01-04
## 1: 12.23777 57.45136
## 2: 12.23777 57.45136
## 3: 12.23777 57.45136
## 4: 12.23777 57.45136
## 5: 12.23777 57.45136
## 6: 12.23777 57.45136
## precip 2015-01-04 - 2015-01-11 precip 2015-01-11 - 2015-01-18
## 1: 30.12705 30.44947
## 2: 30.12705 30.44947
## 3: 30.12705 30.44947
## 4: 30.12705 30.44947
## 5: 30.12705 30.44947
## 6: 30.12705 30.44947
## precip 2015-01-18 - 2015-01-25 precip 2015-01-25 - 2015-02-01
## 1: 1.521829 29.39
## 2: 1.521829 29.39
## 3: 1.521829 29.39
## 4: 1.521829 29.39
## 5: 1.521829 29.39
## 6: 1.521829 29.39
## precip 2015-02-01 - 2015-02-08 precip 2015-02-08 - 2015-02-15
## 1: 32.87832 8.179804
## 2: 32.87832 8.179804
## 3: 32.87832 8.179804
## 4: 32.87832 8.179804
## 5: 32.87832 8.179804
## 6: 32.87832 8.179804
## precip 2015-02-15 - 2015-02-22 precip 2015-02-22 - 2015-03-01
## 1: 0.9639814 16.6591
## 2: 0.9639814 16.6591
## 3: 0.9639814 16.6591
## 4: 0.9639814 16.6591
## 5: 0.9639814 16.6591
## 6: 0.9639814 16.6591
## precip 2015-03-01 - 2015-03-08 precip 2015-03-08 - 2015-03-15
## 1: 3.304466 0
## 2: 3.304466 0
## 3: 3.304466 0
## 4: 3.304466 0
## 5: 3.304466 0
## 6: 3.304466 0
## precip 2019-01-20 - 2019-01-27 precip 2019-01-27 - 2019-02-03
## 1: 12.99262 4.582856
## 2: 12.99262 4.582856
## 3: 12.99262 4.582856
## 4: 12.99262 4.582856
## 5: 12.99262 4.582856
## 6: 12.99262 4.582856
## precip 2019-02-03 - 2019-02-10 precip 2019-02-10 - 2019-02-17
## 1: 35.03753 4.796012
## 2: 35.03753 4.796012
## 3: 35.03753 4.796012
## 4: 35.03753 4.796012
## 5: 35.03753 4.796012
## 6: 35.03753 4.796012
## precip 2019-02-17 - 2019-02-24 precip 2019-02-24 - 2019-03-03
## 1: 28.08331 0
## 2: 28.08331 0
## 3: 28.08331 0
## 4: 28.08331 0
## 5: 28.08331 0
## 6: 28.08331 0
## precip 2019-03-03 - 2019-03-10 precip 2019-03-10 - 2019-03-17
## 1: 58.36246 18.26469
## 2: 58.36246 18.26469
## 3: 58.36246 18.26469
## 4: 58.36246 18.26469
## 5: 58.36246 18.26469
## 6: 58.36246 18.26469
## precip 2019-03-17 - 2019-03-24 precip 2019-03-24 - 2019-03-31
## 1: 17.53749 0.8963228
## 2: 17.53749 0.8963228
## 3: 17.53749 0.8963228
## 4: 17.53749 0.8963228
## 5: 17.53749 0.8963228
## 6: 17.53749 0.8963228
## precip 2019-03-31 - 2019-04-07 precip 2019-04-07 - 2019-04-14
## 1: 1.68 0
## 2: 1.68 0
## 3: 1.68 0
## 4: 1.68 0
## 5: 1.68 0
## 6: 1.68 0
## precip 2019-04-14 - 2019-04-21 precip 2019-04-21 - 2019-04-28
## 1: 0 0
## 2: 0 0
## 3: 0 0
## 4: 0 0
## 5: 0 0
## 6: 0 0
## precip 2019-04-28 - 2019-05-05 precip 2019-05-05 - 2019-05-12
## 1: 0 0
## 2: 0 0
## 3: 0 0
## 4: 0 0
## 5: 0 0
## 6: 0 0
## precip 2019-05-12 - 2019-05-19
## 1: 0
## 2: 0
## 3: 0
## 4: 0
## 5: 0
## 6: 0- X, Y coordinates correspond to longitude and latitude.
0.1 Delimitate the area of the study
lng1 <- min(Train$Longitude)
lng2 <- max(Train$Longitude)
lat1 <- min(Train$Latitude)
lat2 <- max(Train$Latitude)
leaflet(Train) %>%
addTiles() %>%
setView(lng = (lng1+lng2)/2, lat = (lat1+lat2)/2, zoom = 8.4) %>%
addRectangles(
lng1= lng1, lat1= lat1,
lng2= lng2, lat2=lat2,
fillColor = "transparent"
)1 Glimpse to Hydrology, Elevation, Landscover, and Population Maps
Hydrology <- EBImage::readImage("Hydrology_bib.png")
Elevation <- EBImage::readImage("elevation_bib.png")
Landcover <- EBImage::readImage("LandCover_bib.png")
Population <- EBImage::readImage("Population_bib.png")
par(mfrow=c(2,2))
plot(Elevation)
title("Elevation")
plot(Landcover)
title("Landscovers")
plot(Hydrology)
title("Hydrology")
plot(Population)
title("Population")
1.1 Glimpse on the Altitude (Elevation) and Type soil in the region using train dataset
1.1.1 Soil Description
- The Color palette and the description of the Soil Name are loaded from this link.
Soil_pal <- fread("LC_Type1_mode.csv")
#LC_type1_palette <- c("#05450a", "#086a10", "#54a708", "#78d203", "#009900", "#c6b044", "#dcd159", "#dade48", "#fbff13", "#b6ff05", "#27ff87", "#c24f44", "#a5a5a5", "#ff6d4c", "#69fff8", "#f9ffa4", "#1c0dff")
Soil_pal <- Soil_pal %>%
rename(LC_Type1_mode = Value) %>%
separate(Description, into = c("Soil Name", "Description"), sep = ":" , remove = FALSE) %>%
mutate(`Soil Name` = as.factor(`Soil Name`))
Soil_pal## LC_Type1_mode Color Soil Name
## 1 1 #05450a Evergreen Needleleaf Forests
## 2 2 #086a10 Evergreen Broadleaf Forests
## 3 3 #54a708 Deciduous Needleleaf Forests
## 4 4 #78d203 Deciduous Broadleaf Forests
## 5 5 #009900 Mixed Forests
## 6 6 #c6b044 Closed Shrublands
## 7 7 #dcd159 Open Shrublands
## 8 8 #dade48 Woody Savannas
## 9 9 #fbff13 Savannas
## 10 10 #b6ff05 Grasslands
## 11 11 #27ff87 Permanent Wetlands
## 12 12 #c24f44 Croplands
## 13 13 #a5a5a5 Urban and Built-up Lands
## 14 14 #ff6d4c Cropland/Natural Vegetation Mosaics
## 15 15 #69fff8 Permanent Snow and Ice
## 16 16 #f9ffa4 Barren
## 17 17 #1c0dff Water Bodies
## Description
## 1 dominated by evergreen conifer trees (canopy >2m). Tree cover >60%.
## 2 dominated by evergreen broadleaf and palmate trees (canopy >2m). Tree cover >60%.
## 3 dominated by deciduous needleleaf (larch) trees (canopy >2m). Tree cover >60%.
## 4 dominated by deciduous broadleaf trees (canopy >2m). Tree cover >60%.
## 5 dominated by neither deciduous nor evergreen (40-60% of each) tree type (canopy >2m). Tree cover >60%.
## 6 dominated by woody perennials (1-2m height) >60% cover.
## 7 dominated by woody perennials (1-2m height) 10-60% cover.
## 8 tree cover 30-60% (canopy >2m).
## 9 tree cover 10-30% (canopy >2m).
## 10 dominated by herbaceous annuals (<2m).
## 11 permanently inundated lands with 30-60% water cover and >10% vegetated cover.
## 12 at least 60% of area is cultivated cropland.
## 13 at least 30% impervious surface area including building materials, asphalt and vehicles.
## 14 mosaics of small-scale cultivation 40-60% with natural tree, shrub, or herbaceous vegetation.
## 15 at least 60% of area is covered by snow and ice for at least 10 months of the year.
## 16 at least 60% of area is non-vegetated barren (sand, rock, soil) areas with less than 10% vegetation.
## 17 at least 60% of area is covered by permanent water bodies.## Color palette of Land Cover Type 1 (LC_Type1_mode)
require(scales)
scales::show_col(Soil_pal$Color)
# preserve the order of levels as in the Soil_pal dataframe
#levels(Soil_pal$`Soil Name`) <- Soil_pal$`Soil Name`
#Create a custom color scale
require(RColorBrewer)
colors <- Soil_pal$Color
names(colors) <- Soil_pal$`Soil Name`
colScale <- scale_colour_manual(name = "Soil Name", values = colors)
as.data.frame(colors)## colors
## Evergreen Needleleaf Forests #05450a
## Evergreen Broadleaf Forests #086a10
## Deciduous Needleleaf Forests #54a708
## Deciduous Broadleaf Forests #78d203
## Mixed Forests #009900
## Closed Shrublands #c6b044
## Open Shrublands #dcd159
## Woody Savannas #dade48
## Savannas #fbff13
## Grasslands #b6ff05
## Permanent Wetlands #27ff87
## Croplands #c24f44
## Urban and Built-up Lands #a5a5a5
## Cropland/Natural Vegetation Mosaics #ff6d4c
## Permanent Snow and Ice #69fff8
## Barren #f9ffa4
## Water Bodies #1c0dffp1 <- Train %>%
distinct(Square_ID, .keep_all = TRUE) %>%
ggplot() +
aes(x = Longitude, y = Latitude, colour = `Altitude (m)`) +
geom_point(size = 5) +
scale_colour_gradientn(colours = terrain.colors(10))
## add box to urban zone
p1bis <- p1 +
geom_rect(aes(xmin = 34.95, xmax = 35.1, ymin = -15.71, ymax =-15.87),
fill = "transparent", color = "black", size = 0.5) +
geom_rect(aes(xmin = 35.05, xmax = 35.15, ymin = -16.52, ymax =-16.64),
fill = "transparent", color = "black", size = 0.5) +
geom_rect(aes(xmin = 34.8, xmax = 34.88, ymin = -15.98, ymax =-16.29),
fill = "transparent", color = "black", size = 0.5) +
geom_rect(aes(xmin = 35.39, xmax = 35.41, ymin = -15.77, ymax =-15.85),
fill = "transparent", color = "black", size = 0.5) +
geom_rect(aes(xmin = 35.38, xmax = 35.4, ymin = -15.24, ymax =-15.3),
fill = "transparent", color = "black", size = 0.5) +
ggtitle("Area Altitude distibution and Urban zone localisation")
p2 <- Train %>%
distinct(Square_ID, .keep_all = TRUE) %>%
left_join(Soil_pal, by = "LC_Type1_mode") %>%
#filter(LC_Type1_mode == 13) %>%
ggplot() +
aes(x = Longitude, y = Latitude, colour = `Soil Name`) +
geom_point(size = 2) +
colScale +
#scale_colour_manual(values = unique(full$Color)) +
geom_rect(aes(xmin = 34.95, xmax = 35.1, ymin = -15.71, ymax =-15.87),
fill = "transparent", color = "black", size = 0.5) +
geom_rect(aes(xmin = 35.05, xmax = 35.15, ymin = -16.52, ymax =-16.64),
fill = "transparent", color = "black", size = 0.5) +
geom_rect(aes(xmin = 34.8, xmax = 34.88, ymin = -15.98, ymax =-16.29),
fill = "transparent", color = "black", size = 0.5) +
geom_rect(aes(xmin = 35.39, xmax = 35.41, ymin = -15.77, ymax =-15.85),
fill = "transparent", color = "black", size = 0.5) +
geom_rect(aes(xmin = 35.38, xmax = 35.4, ymin = -15.24, ymax =-15.3),
fill = "transparent", color = "black", size = 0.5) +
ggtitle("Soil Name distibution and Urban zone localisation")
gridExtra::grid.arrange( p1bis, p2, ncol = 2)
We obtain similar plots compared to
ElevationandLandcoverplots from biblography.The right plot shows less Altitude of the green dark area in the south of the Malawi.
The right plot shows the Soil Name distribution. The Urban zone is indicated in grey color. The most large urban zone is the center of the region, loacted in
CroplandsSoil Name and in relative high altitude (more that 1000 m (yellow)). The four other urban zone are smaller are located in green area which can more exposed to flood.
1.2 Omit text and axis from ggplot
omit <- theme_bw() + theme(panel.border = element_blank(), panel.grid.major = element_blank(),
panel.grid.minor = element_blank(), axis.line = element_blank(), axis.text.x=element_blank(),axis.text.y=element_blank(), axis.ticks.x=element_blank(), axis.ticks.y=element_blank(), axis.title.x=element_blank(), axis.title.y=element_blank(), legend.title = element_blank(), legend.position = "none", plot.title =element_blank())
## omit all text, axes background from plot for image segmentation if any
soil <- Train %>%
ggplot() +
aes(x = Longitude, y = Latitude, colour = `Altitude (m)`) +
geom_point(size = 6) +
scale_colour_gradient(low = "darkblue", high = "white")+
omit
# Save image
invisible(ggsave("Soil.jpeg", plot = soil, dpi = 300))## Saving 14 x 8 in image1.3 Preprocessing Train data
new_train <- Train %>%
gather(key = `Week of` , value = Pluviometry, 8:ncol(Train)) %>%
mutate(`Week of` = str_extract(`Week of`, "\\d{4}-\\d{2}-\\d{2}") ) %>%
mutate(`Week of` = as.Date(`Week of`, format= "%Y-%m-%d")) %>%
mutate(Week = lubridate::week(`Week of`)) %>%
mutate(Year = if_else(`Week of` <= "2015-03-15", 2015, 2019)) %>%
group_by(Square_ID, Year) %>%
mutate(cum_Pluvio = cumsum(Pluviometry)) %>%
ungroup() %>%
mutate(Target = if_else(Year == 2015, target_2015, 0)) %>%
left_join(Soil_pal, by = "LC_Type1_mode") %>%
rename(Soil_type = LC_Type1_mode) %>%
mutate(`Target Range` = cut(Target, breaks= c(-Inf, 0, 0.25, 0.5, 0.75, 1) ,
labels=c("0", "Low" , "Middle" , "High", "Sure"))) %>%
mutate(Height = cut(`Altitude (m)`, breaks= c( 44, 100 ,400, 650 ,750, 1000 , 2300 , 2804) ,
labels=c("R1","L1" , "L2" , "L3", "L4", "L5", "L6"))) %>%
select(-target_2015)
new_train %>% head## # A tibble: 6 x 17
## Longitude Latitude `Altitude (m)` Soil_type Square_ID other `Week of`
## <dbl> <dbl> <dbl> <int> <chr> <chr> <date>
## 1 34.3 -15.9 888. 9 4e3c3896 14ce… 2014-11-16
## 2 34.3 -15.9 743. 9 4e3c3897 14ce… 2014-11-16
## 3 34.3 -15.9 566. 9 4e3c3898 14ce… 2014-11-16
## 4 34.3 -15.9 443. 10 4e3c3899 14ce… 2014-11-16
## 5 34.3 -15.9 437. 10 4e3c389a 14ce… 2014-11-16
## 6 34.3 -15.9 406. 10 4e3c389b 14ce… 2014-11-16
## # … with 10 more variables: Pluviometry <dbl>, Week <dbl>, Year <dbl>,
## # cum_Pluvio <dbl>, Target <dbl>, Color <chr>, `Soil Name` <fct>,
## # Description <chr>, `Target Range` <fct>, Height <fct>1.4 Set Water bodies, Permanent Wetland and Rivers areas with independent class even that have the same Altitude with other Soil Name
new_train <- new_train %>%
mutate(Height = cut(`Altitude (m)`, breaks= c( 44, 100 ,465,480, 700, 715,750, 1000 , 2300 , 2804) ,
labels=c("River1","L1" ,"River2" ,"L2" , "L3","L4", "L5", "L6", "L7"))) %>%
mutate(Height = as.character(Height)) %>%
mutate(Height = if_else(`Soil Name` == "Water Bodies", "Water", Height)) %>%
mutate(Height = if_else(`Soil Name` == "Permanent Wetlands", "Wetlands", Height))%>%
mutate(Height= if_else(Height == "L2" & Target >= 0.7, "River3", Height))
## Add River3 to Height L2 for 2019
SquareID_River3 <- new_train %>% filter(Year == 2015) %>% filter(Height == "River3") %>% distinct(Square_ID, .keep_all = FALSE)
new_train <-new_train %>%
mutate(Height = if_else(Year == 2019 & Height == "L2" & Square_ID %in% SquareID_River3$Square_ID, "River3", Height)) 1.5 Visualize Pluviometry
new_train %>%
ggplot() +
geom_line( aes(x = `Week of`, y = Pluviometry), color = "blue")+
geom_line(aes(x = `Week of`, y = cum_Pluvio), color = "red")+
geom_smooth(aes(x = `Week of`, y = cum_Pluvio), method = "gam", formula = y ~ s(x, bs = "cs"), color = "black")+
ylim(0, 400) +
facet_wrap(Year ~., ncol = 2,scales = "free")
1.6 Which Soil is more in flood?
new_train %>%
filter(Year == 2015) %>%
mutate(wrap_mode = as.factor(if_else(Target == 0, "No Risk 2015", "With Risk 2015"))) %>%
group_by(Soil_type, Height, `Soil Name`, wrap_mode)%>%
summarise(Frequency = log(n())) %>%
ungroup() %>%
mutate(Soil_Height = as.factor(paste0( Height, Soil_type))) %>%
ggplot() +
aes(x = Soil_Height, y= Frequency, fill = `Soil Name`) +
geom_col() +
facet_grid(wrap_mode ~ .)+
theme(legend.position="bottom") 
1.7 Weight for Soil/Height/Flood
SoilHeight_weight <- new_train %>%
#filter(Year == 2015) %>%
mutate(wrap_mode = as.factor(if_else(Target == 0, "No Risk 2015", "With Risk 2015"))) %>%
group_by(Soil_type, Height, `Soil Name`, wrap_mode)%>%
summarise(Frequency = log(n()))%>%
ungroup() %>%
mutate(Soil_Height = as.factor(paste0( Height, Soil_type))) %>%
group_by(Soil_Height) %>%
mutate(SumFreq= sum(Frequency)) %>%
#group_by(wrap_mode, add=TRUE) %>%
mutate(SoilHeight_Weight=Frequency/SumFreq) %>%
ungroup() %>%
select(Height, Soil_type, SoilHeight_Weight, wrap_mode ) %>%
arrange(Height)
new_train <- new_train %>%
mutate(wrap_mode = as.factor(if_else(Target == 0, "No Risk 2015", "With Risk 2015"))) %>%
left_join(SoilHeight_weight, by = c("Height", "Soil_type", "wrap_mode"))
new_train %>% head## # A tibble: 6 x 19
## Longitude Latitude `Altitude (m)` Soil_type Square_ID other `Week of`
## <dbl> <dbl> <dbl> <int> <chr> <chr> <date>
## 1 34.3 -15.9 888. 9 4e3c3896 14ce… 2014-11-16
## 2 34.3 -15.9 743. 9 4e3c3897 14ce… 2014-11-16
## 3 34.3 -15.9 566. 9 4e3c3898 14ce… 2014-11-16
## 4 34.3 -15.9 443. 10 4e3c3899 14ce… 2014-11-16
## 5 34.3 -15.9 437. 10 4e3c389a 14ce… 2014-11-16
## 6 34.3 -15.9 406. 10 4e3c389b 14ce… 2014-11-16
## # … with 12 more variables: Pluviometry <dbl>, Week <dbl>, Year <dbl>,
## # cum_Pluvio <dbl>, Target <dbl>, Color <chr>, `Soil Name` <fct>,
## # Description <chr>, `Target Range` <fct>, Height <chr>, wrap_mode <fct>,
## # SoilHeight_Weight <dbl>1.8 Categorize elevation to set slopes and define how water moves in the environment
new_train %>%
filter(Year == 2015) %>%
distinct(Square_ID, .keep_all = TRUE) %>%
ggplot +
geom_point(aes(x = Longitude, y = Latitude, color = Height)) + #
geom_point(data = subset(new_train, Target > 0), aes(x = Longitude, y = Latitude,
colour = `Target Range`, shape = `Target Range`),
size = 0.8, stroke = 0, shape = 16) +
#scale_shape_manual(values=c(3, 16, 17, 23), labels = c("0.25", "0.5", "0.75", "1")) +
scale_colour_manual(values = c( "blue" ,"green" , "lightgreen" , "pink" ,"gold" ,"orange" , "darkorange1" ,"chocolate", "cyan" ,"lightblue" , "darkseagreen1","darkseagreen1","darkorchid", "red" , "azure2","burlywood1"),
labels = c("0.75", "L1", "L2", "L3", "L4","L5", "L6", "L7" , "0.25", "0.5","River1", "River2","River3", "1", "Water" , "Wetlands" ),name = "Height/Target") +
ggtitle("Map Altitudes and Flood Risks: 7 Height levels and 4 Probabilities range risks") 
1.9 Target distribution
new_train %>%
filter(Year == 2015) %>%
distinct(Square_ID, .keep_all = TRUE) %>%
mutate(wrap_mode = as.factor(if_else(Target == 0, "No Risk 2015", "With Risk 2015"))) %>%
group_by(`Target Range`, wrap_mode) %>%
summarise(Frequency = n()) %>%
ungroup() %>%
ggplot +
aes(x = `Target Range`, y = Frequency, colour = `Target Range`) +
geom_point()+
facet_wrap(wrap_mode~ ., ncol = 2 , scales = "free") +
scale_colour_manual(values = terrain.colors(6),labels = c( 0, 0.25, 0.5, 0.75, 1))+
ggtitle("Flood Risk Distribution") 
We note:
Unbalanced dataset by class
0without risk. We can reduce the size of class0, by omittingWater bodyarea, and highest Altitude (>2300m).The surface with High risk range [0.75,1] is the lowest class, followed by Middle with probability between [0.5, 0.75].
The proportion of area with no risk coverts the most important surface during the Flood 2015.
We expect that these values will increase for surfaces with risk, if the pluviometry is higher during flood 2019.
2 Deal with unbalanced dataset
# balanced_train <- new_train %>%
# filter(Year == 2015) %>%
# #filter(Target== 0) %>%
# group_by(`Target Range`, Height) %>%
# sample_n(1000, replace = TRUE) %>%
# ungroup() %>%
# select(Longitude, Latitude, `Target Range`, `Altitude (m)`, Soil_type, Pluviometry, cum_Pluvio, Target, Height, SoilHeight_Weight)
# #filter(Target > 0) %>%
# # group_by(Height) %>%
# #summarise(n())
#
#
# balanced_train %>%
# mutate(wrap_mode = as.factor(if_else(Target == 0, "No Risk 2015", "With Risk 2015"))) %>%
# group_by(`Target Range`, wrap_mode) %>%
# summarise(Frequency = n()) %>%
# ungroup() %>%
# ggplot +
# aes(x = `Target Range`, y = Frequency, colour = `Target Range`) +
# geom_point()+
# facet_wrap(wrap_mode~ ., ncol = 2 , scales = "free") +
# scale_colour_manual(values = terrain.colors(6),labels = c( 0, 0.25, 0.5, 0.75, 1))+
# ggtitle("Flood Risk Distribution (Balanced)") 2.1 Flood Zone and Pluviomtery: 2015 versus 2019
Square_ID_F <- new_train %>%
filter(`Week of` <= "2015-03-15") %>%
#filter(Target > 0) %>%
select(Square_ID) %>%
unique()
new_train %>%
filter(Square_ID %in% Square_ID_F$Square_ID) %>%
group_by(Year)%>%
filter(`Week of` == max(`Week of`))%>%
ggplot() +
aes(x = Longitude, y = Latitude) +
geom_point(aes(colour = cum_Pluvio, size = Target)) + # , size = target_2015
scale_colour_gradientn(colours = c("red","darkblue","blue","lightblue","white"),
values = c(1.0,0.8,0.6,0.4,0.2,0)) +
#geom_point(data = subset(new_train, is.na(Target)), aes(size = 0, colour = cum_Pluvio)) +
scale_size(name = "Target",
breaks = c(0, 0.5, 1, 2)) +
facet_wrap(Year ~., ncol = 2, scales = "free") +
geom_rect(aes(xmin = 34.95, xmax = 35.1, ymin = -15.71, ymax =-15.87),
fill = "transparent", color = "black", size = 0.5) +
geom_rect(aes(xmin = 35.05, xmax = 35.15, ymin = -16.52, ymax =-16.64),
fill = "transparent", color = "black", size = 0.5) +
geom_rect(aes(xmin = 34.8, xmax = 34.88, ymin = -15.98, ymax =-16.29),
fill = "transparent", color = "black", size = 0.5) +
geom_rect(aes(xmin = 35.39, xmax = 35.41, ymin = -15.77, ymax =-15.85),
fill = "transparent", color = "black", size = 0.5) +
geom_rect(aes(xmin = 35.38, xmax = 35.4, ymin = -15.24, ymax =-15.3),
fill = "transparent", color = "black", size = 0.5) +
ggtitle("Flood Zone during 2015 and Cumulative Pluviometry Maps ")
Depending only on cumulative pluviometry we expect to have more flood surface during 2019 compared to flood during 2015.
Cumulative pluviomtery of the same zone during 2019 is higher (2 times) that those of 2015.
The Flood seems to be more invasive and the water will reach higher altitude and more area.
3 Random Forest modeling of Flood Zone Probability
# library(randomForest)
# train_2015 <- balanced_train %>%
# #filter(Year == 2015) %>%
# rename(Altitude = `Altitude (m)`) %>%
# mutate(Altitude = log(Altitude)) %>%
# mutate(Height = as.numeric(as.factor(Height))) %>%
# select(Longitude, Latitude, Altitude, Target, Soil_type, cum_Pluvio, Pluviometry, Height, SoilHeight_Weight)
#
#
# rf_model <- randomForest(Target ~ Longitude +Latitude + Soil_type + cum_Pluvio + Altitude + Pluviometry + Height + SoilHeight_Weight +
# Target_Range ,
# data = train_2015,
# ntree=500)
#
# load("rf_model.RData")
# save(rf_model, file ="rf_model.RData")
#
# summary(rf_model)The flood zone probability has high correlation with
Cumulative Pluviometry,Soil Name, andAltitude.Predict Flood zone probability during 2019.
# train_2019 <- new_train %>%
# filter(Year == 2019) %>%
# rename(Altitude = `Altitude (m)`) %>%
# mutate(Altitude = log(Altitude)) %>%
# mutate(Height = as.numeric(as.factor(Height))) %>%
# select(Longitude, Latitude,Altitude, -Target, Soil_type, cum_Pluvio, Pluviometry, Height, SoilHeight_Weight)
# # select(-Target, Altitude, Soil_type, cum_Pluvio, Pluviometry)
#
# target_2019 <- round(predict(rf_model, train_2019, type="response"), digits = 2)
#
#
#
# summary(target_2019)
#
#
# new_train[new_train$Year == 2019,]$Target <- target_20193.1 Visualize RF Prediction
# new_train %>%
# ggplot() +
# aes(x = Longitude, y = Latitude) + #, colour = cum_Pluvio
# geom_point(aes(colour = Target, size = Target)) +
# #scale_colour_gradient(low = "blue", high = "red") +
# scale_colour_gradientn(colours = c("red","darkblue","blue","lightblue","white"),
# values = c(1.0,0.75,0.5,0.25,0)) +
# scale_size(name = "Target",
# breaks = c(0, 0.25, 0.5, 0.75, 1)) +
# facet_wrap(Year ~., ncol = 2) +
# geom_rect(aes(xmin = 34.95, xmax = 35.1, ymin = -15.71, ymax =-15.87),
# fill = "transparent", color = "black", size = 0.5) +
# geom_rect(aes(xmin = 35.05, xmax = 35.15, ymin = -16.52, ymax =-16.64),
# fill = "transparent", color = "black", size = 0.5) +
# geom_rect(aes(xmin = 34.8, xmax = 34.88, ymin = -15.98, ymax =-16.29),
# fill = "transparent", color = "black", size = 0.5) +
# geom_rect(aes(xmin = 35.39, xmax = 35.41, ymin = -15.77, ymax =-15.85),
# fill = "transparent", color = "black", size = 0.5) +
# geom_rect(aes(xmin = 35.38, xmax = 35.4, ymin = -15.24, ymax =-15.3),
# fill = "transparent", color = "black", size = 0.5)+
# ggtitle("Flood zone prediction (RF) during 2019: Prob = ft(Cumulative Pluviometry, Altitude, Pluviometry, Soil_type)")Random Forest prediction does not give a good predction. We expect to have larger flood zone than flood 2015.
The disaster seems to be more invasive and mostly all the south will be flooded except zone with Altitude more than 2000 m
4 Caret
Very time consuming!
require(caret)
# set.seed(42)
# index <- createDataPartition(train_2015$Target, p = 0.7, list = FALSE)
# train_data <- train_2015[index, ]
# test_data <- train_2015[-index, ]
#
# set.seed(42)
# model_rf <- caret::train(Target ~ .,
# data = train_data,
# method = "rf",
# metric = "RMSE",
# preProcess = c("scale", "center"),
# trControl = trainControl(method = "repeatedcv",
# number = 2,
# repeats = 1,
# verboseIter = TRUE)
# )
#
#
# save(model_rf, file ="model_caret_rf.RData")
#
# final <- data.frame(actual = as.factor(as.character(test_data$Target)),
# pred = predict(model_rf, newdata = test_data, type = "raw"))
#
# #final$predict <- as.factor(ifelse(final$pred >= 0.75, 1, 0))
#
# #cm_original <- confusionMatrix(final$predict, final$actual)
#
# target_2019 <- round(predict(model_rf, train_2019, type="raw"), digits = 2)
#
#
#
# summary(target_2019)
#
#
# #new_train[new_train$Year == 2019,]$Target <- target_20195 Round Target to (0,1)
new_train01 <- new_train %>%
mutate(Target = as.factor(if_else(Target >= 0.5, 1, 0)))%>%
group_by(Square_ID) %>%
mutate(meanPluv = mean(Pluviometry),
medianPluv = median(Pluviometry),
maxPluv = max(Pluviometry))%>%
ungroup()
new_train01 %>%
filter(Year == 2015)%>%
group_by(Target) %>%
summarise(nbr = n()) %>%
ggplot() +
aes(x = Target, y = nbr, fill= Target) +
geom_col()
new_train01 %>%
#mutate(Target = as.factor(Target)) %>%
filter(Year == 2015) %>%
ggplot +
geom_point(aes(x = Longitude, y = Latitude, color = Height)) +
geom_point(#data = subset(new_train, Target == 1),
aes(x = Longitude, y = Latitude, color = Target, shape = Target),
size = 0.6, stroke = 0, shape = 19) +
scale_colour_manual(values = c( "lightblue" , "red" , "green" , "lightgreen" , "pink" ,"gold" ,"orange" , "darkorange1" ,"chocolate", "darkseagreen1","darkseagreen1","darkorchid" , "azure2","burlywood1"),
labels = c( "0", "1", "L1", "L2", "L3", "L4","L5", "L6", "L7" , "River1", "River2","River3", "Water" , "Wetlands" ), name = "Height/Target") 
5.1 Add target Range variable to 2019
Target_Range_vs_XY_2015 <- new_train01 %>%
filter(Year == 2015) %>%
distinct(Square_ID, .keep_all = TRUE) %>%
select(Square_ID,`Target Range`)
Train_2019 <- new_train01 %>%
filter(Year == 2019) %>%
select(-`Target Range`)
Train_2019_TR <- Train_2019 %>%
left_join(Target_Range_vs_XY_2015, by = "Square_ID")
new_train01[new_train01$Year == 2019,]$`Target Range` <- Train_2019_TR$`Target Range`
new_train01[new_train01$Year == 2019,] %>% head## # A tibble: 6 x 22
## Longitude Latitude `Altitude (m)` Soil_type Square_ID other `Week of`
## <dbl> <dbl> <dbl> <int> <chr> <chr> <date>
## 1 34.3 -15.9 888. 9 4e3c3896 14ce… 2019-01-20
## 2 34.3 -15.9 743. 9 4e3c3897 14ce… 2019-01-20
## 3 34.3 -15.9 566. 9 4e3c3898 14ce… 2019-01-20
## 4 34.3 -15.9 443. 10 4e3c3899 14ce… 2019-01-20
## 5 34.3 -15.9 437. 10 4e3c389a 14ce… 2019-01-20
## 6 34.3 -15.9 406. 10 4e3c389b 14ce… 2019-01-20
## # … with 15 more variables: Pluviometry <dbl>, Week <dbl>, Year <dbl>,
## # cum_Pluvio <dbl>, Target <fct>, Color <chr>, `Soil Name` <fct>,
## # Description <chr>, `Target Range` <fct>, Height <chr>, wrap_mode <fct>,
## # SoilHeight_Weight <dbl>, meanPluv <dbl>, medianPluv <dbl>, maxPluv <dbl>5.2 Save Train2015 and test2019 dataset for Python
dataset2019 <- new_train01 %>%
select(Longitude, Latitude,`Week of` , Week, `Altitude (m)`,
Soil_type, Square_ID, Pluviometry, cum_Pluvio, Target,
`Target Range`, Height, SoilHeight_Weight, meanPluv, medianPluv, maxPluv) %>%
rename(weeks = `Week of`) %>%
filter(weeks > "2015-03-30") %>%
rename(Altitude = `Altitude (m)`) %>%
mutate(Altitude = log(Altitude)) %>%
rename(Target_Range = `Target Range`) %>%
mutate(Target_Range = as.numeric(Target_Range) -1) %>%
mutate(Height = as.numeric(as.factor(Height)) - 1)
fwrite(dataset2019, "dataset2019.csv")
head(dataset2019)## # A tibble: 6 x 16
## Longitude Latitude weeks Week Altitude Soil_type Square_ID Pluviometry
## <dbl> <dbl> <date> <dbl> <dbl> <int> <chr> <dbl>
## 1 34.3 -15.9 2019-01-20 3 6.79 9 4e3c3896 13.0
## 2 34.3 -15.9 2019-01-20 3 6.61 9 4e3c3897 13.0
## 3 34.3 -15.9 2019-01-20 3 6.34 9 4e3c3898 13.0
## 4 34.3 -15.9 2019-01-20 3 6.09 10 4e3c3899 13.0
## 5 34.3 -15.9 2019-01-20 3 6.08 10 4e3c389a 13.0
## 6 34.3 -15.9 2019-01-20 3 6.01 10 4e3c389b 13.0
## # … with 8 more variables: cum_Pluvio <dbl>, Target <fct>, Target_Range <dbl>,
## # Height <dbl>, SoilHeight_Weight <dbl>, meanPluv <dbl>, medianPluv <dbl>,
## # maxPluv <dbl>6 Animated Pluviometry of 2015 and 2019
new_train %>%
filter(`Week of` <= "2015-04-15") %>%
ggplot() +
aes(x = Longitude, y = Latitude, colour = cum_Pluvio) +
geom_point(size = 4) +
#geom_point( aes(x = Longitude, y = Latitude, colour = Target))+ # data = subset(new_train[new_train$Target > 0.75,]),
#scale_colour_gradient(low = "blue", high = "red") +
#scale_colour_gradientn(colours = rainbow(3))+
scale_colour_gradientn(colours = c("red","darkblue","blue","lightblue","white"),
values = c(1.0,0.8,0.6,0.4,0.2,0)) +
#scale_colour_gradientn(colours = c("blue","lightblue","white"),
# values = c(1.0,0.5,0)) +
facet_wrap(`Week of` ~., ncol = 4 ) +
ggtitle("Cumulative Pluviometry of 2015 before Flood") 
new_train %>%
filter(`Week of` > "2015-03-15") %>%
ggplot() +
aes(x = Longitude, y = Latitude, colour = cum_Pluvio) +
geom_point(size = 4) +
#scale_colour_gradient(low = "blue", high = "red") +
scale_colour_gradientn(colours = c("red","darkblue","blue","lightblue","white"),
values = c(1.0,0.8,0.6,0.4,0.2,0)) +
facet_wrap(`Week of` ~., ncol = 4 ) +
ggtitle("Cumulative Pluviometry of 2019 before Flood") 
The train (2014-2015) and the test (2019) have the same period (17 weeks) but not the same season (2014-Nov : 2015-March, Jan-2019 : May-2019).
Pluviometry during 2019 is two times more that pluviometry during 2015.
6.1 Google Earth Engine Precipitation view
We used Google Earth Engine to see pluvimmetry before Flood 2015 and 2019 around the south of Malawi.
6.2 Get animated Cumulative Pluviometery
# require(gganimate)
#
# goo <- new_train %>%
# filter(`Week of` <= "2015-03-15") %>%
# ggplot() +
# aes(x = Longitude, y = Latitude, colour = cum_Pluvio) +
# geom_point(size = 4) +
# #scale_colour_gradient(low = "blue", high = "red") +
# scale_colour_gradientn(colours = c("blue","lightblue","white"),
# values = c(1.0,0.5,0)) +
# #facet_wrap(`Week of` ~., ncol = 3 ) +
# transition_time(`Week of`) +
# labs(title = "Day: {frame_time}")
#
# # foo <- new_train %>%
# # filter(`Week of` > "2015-03-15") %>%
# # ggplot() +
# # aes(x = Longitude, y = Latitude, colour = Pluviometry) +
# # geom_point(size = 4) +
# # #facet_wrap(`Week of` ~., ncol = 3 ) +
# # transition_time(`Week of`) +
# # labs(title = "Day: {frame_time}")
#
#
# ## get and save animation
# nframes <- length(unique(goo$data$`Week of`)) * 7
# ani_goo <- animate(goo, nframes = nframes, fps = 10)
# png_files <- list.files(".", pattern = ".*png$", full.names = TRUE)
# require(gifski)
# #gifski(png_files, gif_file = "ani_goo.gif", width = 800, height = 600, delay = 1)
# #invisible(file.remove(png_files, ))
#
# #animate(goo, renderer = ffmpeg_renderer(format = "webm"))
# #animate(foo, renderer = ffmpeg_renderer(format = "webm"))# require(plotly)
#
# new_train %>%
# filter(`Week of` < "2014-12-14") %>%
# plot_ly(
# x = ~Longitude,
# y = ~Latitude,
# #size = ~pop,
# color = ~Pluviometry,
# frame = ~ as.Date(`Week of`),
# text = ~`Week of`,
# hoverinfo = "text",
# type = 'scatter',
# mode = 'markers'
# ) %>%
# layout(
# xaxis = list(
# type = "log"
# )
# )7 Xgboost regression with Python
import matplotlib
import matplotlib.pyplot as plt
import pandas as pd
import numpy as np
import seaborn as sns
from sklearn.model_selection import train_test_split
from collections import Counter## Longitude Latitude weeks ... meanPluv medianPluv maxPluv
## 0 34.26 -15.91 2014-11-16 ... 12.788037 4.689434 58.362456
## 1 34.26 -15.90 2014-11-16 ... 12.788037 4.689434 58.362456
## 2 34.26 -15.89 2014-11-16 ... 12.788037 4.689434 58.362456
## 3 34.26 -15.88 2014-11-16 ... 12.788037 4.689434 58.362456
## 4 34.26 -15.87 2014-11-16 ... 12.788037 4.689434 58.362456
##
## [5 rows x 16 columns]## Counter({0: 232917, 1: 19363, 4: 14178, 2: 8024, 3: 5440})
dic_range={0: "0", 1: "Low", 2 : "medium", 3:"High", 4:"Sure"}
plt.show(
plt.bar(range(len(dic_range)),list(Counter(train['Target_Range']).values()))
)
# split Train and Test
# Separate minors and major Target_Range
train0 = train[train.Target_Range == 0]
train1 = train[train.Target_Range == 1]
train2 = train[train.Target_Range == 2]
train3 = train[train.Target_Range == 3]
train4 = train[train.Target_Range == 4]
from sklearn.utils import resample
# Upsample minority classes
train_0_downsampled = resample(train0,
replace= False, # sample with replacement
n_samples=19363, # to match majority class
random_state=123) # reproducible results
#
# Upsample minority classes
#train_1_downsampled = resample(train1,
# replace=False, # sample with replacement
# n_samples=8024, # to match majority class
# random_state=123) # reproducible results
# Downsample majority class
train_2_upsampled = resample(train2,
replace=True, # sample without replacement
n_samples=19363, # to match minority class
random_state=123) # reproducible results
# Downsample majority class
train_3_upsampled = resample(train3,
replace= True, # sample without replacement
n_samples=19363, # to match minority class
random_state=123) #
# Downsample majority class
train_4_upsampled = resample(train4,
replace= True, # sample without replacement
n_samples=19363, # to match minority class
random_state=123) #
# Combine minority class with downsampled majority class
train_balanced = pd.concat([train_0_downsampled, train1, train_2_upsampled, train_3_upsampled, train_4_upsampled])
Counter(train_balanced['Target_Range'])## Counter({0: 19363, 1: 19363, 2: 19363, 3: 19363, 4: 19363})## Index(['Longitude', 'Latitude', 'weeks', 'Week', 'Altitude', 'Soil_type',
## 'Square_ID', 'Pluviometry', 'cum_Pluvio', 'Target', 'Target_Range',
## 'Height', 'SoilHeight_Weight', 'meanPluv', 'medianPluv', 'maxPluv'],
## dtype='object')
ytrain = train.Target
xtrain = train.drop(['Target', 'Square_ID', 'weeks', 'Latitude', 'Longitude', 'meanPluv', 'medianPluv', 'maxPluv', 'Pluviometry'], axis=1) #
xtest = test.drop(['Target', 'weeks', 'Latitude', 'Longitude', 'meanPluv', 'medianPluv', 'maxPluv', 'Pluviometry'], axis=1) #
## Dummy variable levels is not better
# def prepare_data_for_model(raw_dataframe, target_columns, drop_first = True, make_na_col = False):
# # dummy all categorical fields
# dataframe_dummy = pd.get_dummies(raw_dataframe, columns=target_columns,
# drop_first=drop_first,
# dummy_na=make_na_col)
# return (dataframe_dummy)
#
#
# # create dummy features
# xtrain_dum = prepare_data_for_model(xtrain, target_columns=['Soil_type', 'Height']) #, 'Nstage' : 3 classes
# xtrain_dum = xtrain_dum.dropna()
#
# # create dummy features for test
# xtest_dum = prepare_data_for_model(xtest, target_columns=['Soil_type', 'Height']) #, 'Nstage' : 4 classes
# xtest_dum = xtest_dum.dropna()
#
# xtrain_dum.headfrom sklearn.model_selection import train_test_split
import xgboost as xgb
import time
## Split
trn_x, val_x, trn_y, val_y = train_test_split(xtrain , ytrain, random_state = 42, stratify = ytrain, test_size = 0.20) # start_time = time.process_time()
clf = xgb.XGBRegressor(booster = 'gbtree',
objective = 'reg:logistic', # multi:softmax, multi:softprob #'reg:squarederror', # reg:linear reg:logistic binary:logistic
#num_class = 2,
max_depth = 10,
n_estimators = 10000,
min_child_weight = 9,
learning_rate = 0.01,
nthread = 8,
subsample = 0.80,
colsample_bytree = 0.80,
seed = 4242)
clf.fit(trn_x,
trn_y,
eval_set = [(val_x, val_y)],
verbose = True,
#verbose_eval= 10, # print every 10 boost
eval_metric = 'rmse', # rmse, logloss, mae, map, cox-nloglik
early_stopping_rounds = 10)## [0] validation_0-rmse:0.495411
## Will train until validation_0-rmse hasn't improved in 10 rounds.
## [1] validation_0-rmse:0.490456
## [2] validation_0-rmse:0.485552
## [3] validation_0-rmse:0.480697
## [4] validation_0-rmse:0.475898
## [5] validation_0-rmse:0.471438
## [6] validation_0-rmse:0.466731
## [7] validation_0-rmse:0.462064
## [8] validation_0-rmse:0.457445
## [9] validation_0-rmse:0.452875
## [10] validation_0-rmse:0.448646
## [11] validation_0-rmse:0.444613
## [12] validation_0-rmse:0.440168
## [13] validation_0-rmse:0.435771
## [14] validation_0-rmse:0.431419
## [15] validation_0-rmse:0.427409
## [16] validation_0-rmse:0.423144
## [17] validation_0-rmse:0.418916
## [18] validation_0-rmse:0.414731
## [19] validation_0-rmse:0.410867
## [20] validation_0-rmse:0.407207
## [21] validation_0-rmse:0.403134
## [22] validation_0-rmse:0.399486
## [23] validation_0-rmse:0.395487
## [24] validation_0-rmse:0.391545
## [25] validation_0-rmse:0.387644
## [26] validation_0-rmse:0.383777
## [27] validation_0-rmse:0.379935
## [28] validation_0-rmse:0.376431
## [29] validation_0-rmse:0.372664
## [30] validation_0-rmse:0.368958
## [31] validation_0-rmse:0.36527
## [32] validation_0-rmse:0.361621
## [33] validation_0-rmse:0.358279
## [34] validation_0-rmse:0.354709
## [35] validation_0-rmse:0.351181
## [36] validation_0-rmse:0.347669
## [37] validation_0-rmse:0.344599
## [38] validation_0-rmse:0.341162
## [39] validation_0-rmse:0.338015
## [40] validation_0-rmse:0.334644
## [41] validation_0-rmse:0.331641
## [42] validation_0-rmse:0.328325
## [43] validation_0-rmse:0.325307
## [44] validation_0-rmse:0.322071
## [45] validation_0-rmse:0.319188
## [46] validation_0-rmse:0.316008
## [47] validation_0-rmse:0.313231
## [48] validation_0-rmse:0.310111
## [49] validation_0-rmse:0.307018
## [50] validation_0-rmse:0.303959
## [51] validation_0-rmse:0.300929
## [52] validation_0-rmse:0.298188
## [53] validation_0-rmse:0.295218
## [54] validation_0-rmse:0.292277
## [55] validation_0-rmse:0.289362
## [56] validation_0-rmse:0.28648
## [57] validation_0-rmse:0.28387
## [58] validation_0-rmse:0.281039
## [59] validation_0-rmse:0.278241
## [60] validation_0-rmse:0.275478
## [61] validation_0-rmse:0.272731
## [62] validation_0-rmse:0.270015
## [63] validation_0-rmse:0.267325
## [64] validation_0-rmse:0.264663
## [65] validation_0-rmse:0.26203
## [66] validation_0-rmse:0.259647
## [67] validation_0-rmse:0.257297
## [68] validation_0-rmse:0.254734
## [69] validation_0-rmse:0.252203
## [70] validation_0-rmse:0.249691
## [71] validation_0-rmse:0.24748
## [72] validation_0-rmse:0.245012
## [73] validation_0-rmse:0.242575
## [74] validation_0-rmse:0.240159
## [75] validation_0-rmse:0.23777
## [76] validation_0-rmse:0.235399
## [77] validation_0-rmse:0.233056
## [78] validation_0-rmse:0.230737
## [79] validation_0-rmse:0.228712
## [80] validation_0-rmse:0.226696
## [81] validation_0-rmse:0.224442
## [82] validation_0-rmse:0.222511
## [83] validation_0-rmse:0.220295
## [84] validation_0-rmse:0.218406
## [85] validation_0-rmse:0.216453
## [86] validation_0-rmse:0.214603
## [87] validation_0-rmse:0.212471
## [88] validation_0-rmse:0.210353
## [89] validation_0-rmse:0.208262
## [90] validation_0-rmse:0.206188
## [91] validation_0-rmse:0.20414
## [92] validation_0-rmse:0.202103
## [93] validation_0-rmse:0.200094
## [94] validation_0-rmse:0.198103
## [95] validation_0-rmse:0.19634
## [96] validation_0-rmse:0.194642
## [97] validation_0-rmse:0.192709
## [98] validation_0-rmse:0.190789
## [99] validation_0-rmse:0.189109
## [100] validation_0-rmse:0.187226
## [101] validation_0-rmse:0.185365
## [102] validation_0-rmse:0.183518
## [103] validation_0-rmse:0.181696
## [104] validation_0-rmse:0.179884
## [105] validation_0-rmse:0.178099
## [106] validation_0-rmse:0.176528
## [107] validation_0-rmse:0.174773
## [108] validation_0-rmse:0.173033
## [109] validation_0-rmse:0.171314
## [110] validation_0-rmse:0.169611
## [111] validation_0-rmse:0.167922
## [112] validation_0-rmse:0.166251
## [113] validation_0-rmse:0.164599
## [114] validation_0-rmse:0.162962
## [115] validation_0-rmse:0.161339
## [116] validation_0-rmse:0.159735
## [117] validation_0-rmse:0.158369
## [118] validation_0-rmse:0.156982
## [119] validation_0-rmse:0.155419
## [120] validation_0-rmse:0.153876
## [121] validation_0-rmse:0.152346
## [122] validation_0-rmse:0.150833
## [123] validation_0-rmse:0.14933
## [124] validation_0-rmse:0.147845
## [125] validation_0-rmse:0.146378
## [126] validation_0-rmse:0.144921
## [127] validation_0-rmse:0.143478
## [128] validation_0-rmse:0.142261
## [129] validation_0-rmse:0.140846
## [130] validation_0-rmse:0.139446
## [131] validation_0-rmse:0.13806
## [132] validation_0-rmse:0.136687
## [133] validation_0-rmse:0.135504
## [134] validation_0-rmse:0.134156
## [135] validation_0-rmse:0.132821
## [136] validation_0-rmse:0.131705
## [137] validation_0-rmse:0.130395
## [138] validation_0-rmse:0.129271
## [139] validation_0-rmse:0.128158
## [140] validation_0-rmse:0.127062
## [141] validation_0-rmse:0.125798
## [142] validation_0-rmse:0.124549
## [143] validation_0-rmse:0.123484
## [144] validation_0-rmse:0.122256
## [145] validation_0-rmse:0.121215
## [146] validation_0-rmse:0.120011
## [147] validation_0-rmse:0.11882
## [148] validation_0-rmse:0.117636
## [149] validation_0-rmse:0.116468
## [150] validation_0-rmse:0.115309
## [151] validation_0-rmse:0.114167
## [152] validation_0-rmse:0.113033
## [153] validation_0-rmse:0.111911
## [154] validation_0-rmse:0.11096
## [155] validation_0-rmse:0.10986
## [156] validation_0-rmse:0.108766
## [157] validation_0-rmse:0.107686
## [158] validation_0-rmse:0.106779
## [159] validation_0-rmse:0.105915
## [160] validation_0-rmse:0.104862
## [161] validation_0-rmse:0.103823
## [162] validation_0-rmse:0.102791
## [163] validation_0-rmse:0.101771
## [164] validation_0-rmse:0.10076
## [165] validation_0-rmse:0.099917
## [166] validation_0-rmse:0.099109
## [167] validation_0-rmse:0.098284
## [168] validation_0-rmse:0.097308
## [169] validation_0-rmse:0.096342
## [170] validation_0-rmse:0.095387
## [171] validation_0-rmse:0.094439
## [172] validation_0-rmse:0.0935
## [173] validation_0-rmse:0.092574
## [174] validation_0-rmse:0.091657
## [175] validation_0-rmse:0.090747
## [176] validation_0-rmse:0.089847
## [177] validation_0-rmse:0.088954
## [178] validation_0-rmse:0.088071
## [179] validation_0-rmse:0.087198
## [180] validation_0-rmse:0.086332
## [181] validation_0-rmse:0.085477
## [182] validation_0-rmse:0.084628
## [183] validation_0-rmse:0.083789
## [184] validation_0-rmse:0.082956
## [185] validation_0-rmse:0.082298
## [186] validation_0-rmse:0.081482
## [187] validation_0-rmse:0.080673
## [188] validation_0-rmse:0.079871
## [189] validation_0-rmse:0.079081
## [190] validation_0-rmse:0.078297
## [191] validation_0-rmse:0.07768
## [192] validation_0-rmse:0.076908
## [193] validation_0-rmse:0.076146
## [194] validation_0-rmse:0.07539
## [195] validation_0-rmse:0.074641
## [196] validation_0-rmse:0.074031
## [197] validation_0-rmse:0.07343
## [198] validation_0-rmse:0.072702
## [199] validation_0-rmse:0.071981
## [200] validation_0-rmse:0.071399
## [201] validation_0-rmse:0.070844
## [202] validation_0-rmse:0.070142
## [203] validation_0-rmse:0.069446
## [204] validation_0-rmse:0.068758
## [205] validation_0-rmse:0.068076
## [206] validation_0-rmse:0.067401
## [207] validation_0-rmse:0.066732
## [208] validation_0-rmse:0.066069
## [209] validation_0-rmse:0.065415
## [210] validation_0-rmse:0.064765
## [211] validation_0-rmse:0.064122
## [212] validation_0-rmse:0.063627
## [213] validation_0-rmse:0.062995
## [214] validation_0-rmse:0.062371
## [215] validation_0-rmse:0.061753
## [216] validation_0-rmse:0.061139
## [217] validation_0-rmse:0.060534
## [218] validation_0-rmse:0.059934
## [219] validation_0-rmse:0.059474
## [220] validation_0-rmse:0.058885
## [221] validation_0-rmse:0.058301
## [222] validation_0-rmse:0.057722
## [223] validation_0-rmse:0.057283
## [224] validation_0-rmse:0.056833
## [225] validation_0-rmse:0.056269
## [226] validation_0-rmse:0.05571
## [227] validation_0-rmse:0.055159
## [228] validation_0-rmse:0.054612
## [229] validation_0-rmse:0.05407
## [230] validation_0-rmse:0.053534
## [231] validation_0-rmse:0.053003
## [232] validation_0-rmse:0.052477
## [233] validation_0-rmse:0.051956
## [234] validation_0-rmse:0.051442
## [235] validation_0-rmse:0.050931
## [236] validation_0-rmse:0.050426
## [237] validation_0-rmse:0.050051
## [238] validation_0-rmse:0.049554
## [239] validation_0-rmse:0.049063
## [240] validation_0-rmse:0.048681
## [241] validation_0-rmse:0.048324
## [242] validation_0-rmse:0.047844
## [243] validation_0-rmse:0.04737
## [244] validation_0-rmse:0.0469
## [245] validation_0-rmse:0.046435
## [246] validation_0-rmse:0.045975
## [247] validation_0-rmse:0.045519
## [248] validation_0-rmse:0.045067
## [249] validation_0-rmse:0.044721
## [250] validation_0-rmse:0.044278
## [251] validation_0-rmse:0.043838
## [252] validation_0-rmse:0.043405
## [253] validation_0-rmse:0.042974
## [254] validation_0-rmse:0.042548
## [255] validation_0-rmse:0.042126
## [256] validation_0-rmse:0.041709
## [257] validation_0-rmse:0.041295
## [258] validation_0-rmse:0.040885
## [259] validation_0-rmse:0.04048
## [260] validation_0-rmse:0.040078
## [261] validation_0-rmse:0.03968
## [262] validation_0-rmse:0.039287
## [263] validation_0-rmse:0.038988
## [264] validation_0-rmse:0.038601
## [265] validation_0-rmse:0.038219
## [266] validation_0-rmse:0.03784
## [267] validation_0-rmse:0.037465
## [268] validation_0-rmse:0.037093
## [269] validation_0-rmse:0.036813
## [270] validation_0-rmse:0.03655
## [271] validation_0-rmse:0.036189
## [272] validation_0-rmse:0.03583
## [273] validation_0-rmse:0.035561
## [274] validation_0-rmse:0.035306
## [275] validation_0-rmse:0.034957
## [276] validation_0-rmse:0.03461
## [277] validation_0-rmse:0.034267
## [278] validation_0-rmse:0.033927
## [279] validation_0-rmse:0.033677
## [280] validation_0-rmse:0.033445
## [281] validation_0-rmse:0.033113
## [282] validation_0-rmse:0.032785
## [283] validation_0-rmse:0.03246
## [284] validation_0-rmse:0.032138
## [285] validation_0-rmse:0.031819
## [286] validation_0-rmse:0.031504
## [287] validation_0-rmse:0.031191
## [288] validation_0-rmse:0.030883
## [289] validation_0-rmse:0.030577
## [290] validation_0-rmse:0.030274
## [291] validation_0-rmse:0.03006
## [292] validation_0-rmse:0.029843
## [293] validation_0-rmse:0.029629
## [294] validation_0-rmse:0.029424
## [295] validation_0-rmse:0.029133
## [296] validation_0-rmse:0.028844
## [297] validation_0-rmse:0.028558
## [298] validation_0-rmse:0.028355
## [299] validation_0-rmse:0.028156
## [300] validation_0-rmse:0.027877
## [301] validation_0-rmse:0.027682
## [302] validation_0-rmse:0.027408
## [303] validation_0-rmse:0.027136
## [304] validation_0-rmse:0.026867
## [305] validation_0-rmse:0.026601
## [306] validation_0-rmse:0.026417
## [307] validation_0-rmse:0.026155
## [308] validation_0-rmse:0.025981
## [309] validation_0-rmse:0.025804
## [310] validation_0-rmse:0.025548
## [311] validation_0-rmse:0.025295
## [312] validation_0-rmse:0.025044
## [313] validation_0-rmse:0.024875
## [314] validation_0-rmse:0.024719
## [315] validation_0-rmse:0.024475
## [316] validation_0-rmse:0.024323
## [317] validation_0-rmse:0.024083
## [318] validation_0-rmse:0.023844
## [319] validation_0-rmse:0.023686
## [320] validation_0-rmse:0.023452
## [321] validation_0-rmse:0.023303
## [322] validation_0-rmse:0.023072
## [323] validation_0-rmse:0.022843
## [324] validation_0-rmse:0.022618
## [325] validation_0-rmse:0.022394
## [326] validation_0-rmse:0.022252
## [327] validation_0-rmse:0.022032
## [328] validation_0-rmse:0.021814
## [329] validation_0-rmse:0.021684
## [330] validation_0-rmse:0.021469
## [331] validation_0-rmse:0.021331
## [332] validation_0-rmse:0.02112
## [333] validation_0-rmse:0.020911
## [334] validation_0-rmse:0.02079
## [335] validation_0-rmse:0.020584
## [336] validation_0-rmse:0.020454
## [337] validation_0-rmse:0.020331
## [338] validation_0-rmse:0.020129
## [339] validation_0-rmse:0.020009
## [340] validation_0-rmse:0.01989
## [341] validation_0-rmse:0.019693
## [342] validation_0-rmse:0.019573
## [343] validation_0-rmse:0.01938
## [344] validation_0-rmse:0.019188
## [345] validation_0-rmse:0.018998
## [346] validation_0-rmse:0.01881
## [347] validation_0-rmse:0.018624
## [348] validation_0-rmse:0.01844
## [349] validation_0-rmse:0.018257
## [350] validation_0-rmse:0.018077
## [351] validation_0-rmse:0.017898
## [352] validation_0-rmse:0.017721
## [353] validation_0-rmse:0.017546
## [354] validation_0-rmse:0.01744
## [355] validation_0-rmse:0.017268
## [356] validation_0-rmse:0.017097
## [357] validation_0-rmse:0.016995
## [358] validation_0-rmse:0.016895
## [359] validation_0-rmse:0.016729
## [360] validation_0-rmse:0.016631
## [361] validation_0-rmse:0.016467
## [362] validation_0-rmse:0.016304
## [363] validation_0-rmse:0.016212
## [364] validation_0-rmse:0.016052
## [365] validation_0-rmse:0.015893
## [366] validation_0-rmse:0.015736
## [367] validation_0-rmse:0.015649
## [368] validation_0-rmse:0.015494
## [369] validation_0-rmse:0.015341
## [370] validation_0-rmse:0.015257
## [371] validation_0-rmse:0.015181
## [372] validation_0-rmse:0.015031
## [373] validation_0-rmse:0.014882
## [374] validation_0-rmse:0.014735
## [375] validation_0-rmse:0.01459
## [376] validation_0-rmse:0.014445
## [377] validation_0-rmse:0.014302
## [378] validation_0-rmse:0.014233
## [379] validation_0-rmse:0.014092
## [380] validation_0-rmse:0.013953
## [381] validation_0-rmse:0.013877
## [382] validation_0-rmse:0.01374
## [383] validation_0-rmse:0.013604
## [384] validation_0-rmse:0.013469
## [385] validation_0-rmse:0.013336
## [386] validation_0-rmse:0.013264
## [387] validation_0-rmse:0.013133
## [388] validation_0-rmse:0.013003
## [389] validation_0-rmse:0.012874
## [390] validation_0-rmse:0.012807
## [391] validation_0-rmse:0.01274
## [392] validation_0-rmse:0.012673
## [393] validation_0-rmse:0.012548
## [394] validation_0-rmse:0.012424
## [395] validation_0-rmse:0.012359
## [396] validation_0-rmse:0.012237
## [397] validation_0-rmse:0.012175
## [398] validation_0-rmse:0.012055
## [399] validation_0-rmse:0.011936
## [400] validation_0-rmse:0.011818
## [401] validation_0-rmse:0.011701
## [402] validation_0-rmse:0.011585
## [403] validation_0-rmse:0.011471
## [404] validation_0-rmse:0.011357
## [405] validation_0-rmse:0.011301
## [406] validation_0-rmse:0.011246
## [407] validation_0-rmse:0.011135
## [408] validation_0-rmse:0.01108
## [409] validation_0-rmse:0.010971
## [410] validation_0-rmse:0.010862
## [411] validation_0-rmse:0.010755
## [412] validation_0-rmse:0.010703
## [413] validation_0-rmse:0.010597
## [414] validation_0-rmse:0.010493
## [415] validation_0-rmse:0.010389
## [416] validation_0-rmse:0.010286
## [417] validation_0-rmse:0.010185
## [418] validation_0-rmse:0.010084
## [419] validation_0-rmse:0.009984
## [420] validation_0-rmse:0.009886
## [421] validation_0-rmse:0.00984
## [422] validation_0-rmse:0.009742
## [423] validation_0-rmse:0.009697
## [424] validation_0-rmse:0.009601
## [425] validation_0-rmse:0.009506
## [426] validation_0-rmse:0.009412
## [427] validation_0-rmse:0.009319
## [428] validation_0-rmse:0.009276
## [429] validation_0-rmse:0.009185
## [430] validation_0-rmse:0.009094
## [431] validation_0-rmse:0.009004
## [432] validation_0-rmse:0.008915
## [433] validation_0-rmse:0.008827
## [434] validation_0-rmse:0.008787
## [435] validation_0-rmse:0.0087
## [436] validation_0-rmse:0.008614
## [437] validation_0-rmse:0.008529
## [438] validation_0-rmse:0.008444
## [439] validation_0-rmse:0.008361
## [440] validation_0-rmse:0.008278
## [441] validation_0-rmse:0.008197
## [442] validation_0-rmse:0.008116
## [443] validation_0-rmse:0.008036
## [444] validation_0-rmse:0.007956
## [445] validation_0-rmse:0.007877
## [446] validation_0-rmse:0.007843
## [447] validation_0-rmse:0.007765
## [448] validation_0-rmse:0.007731
## [449] validation_0-rmse:0.007655
## [450] validation_0-rmse:0.007579
## [451] validation_0-rmse:0.007504
## [452] validation_0-rmse:0.00743
## [453] validation_0-rmse:0.007357
## [454] validation_0-rmse:0.007284
## [455] validation_0-rmse:0.007253
## [456] validation_0-rmse:0.007222
## [457] validation_0-rmse:0.00715
## [458] validation_0-rmse:0.00712
## [459] validation_0-rmse:0.007096
## [460] validation_0-rmse:0.007026
## [461] validation_0-rmse:0.006957
## [462] validation_0-rmse:0.006888
## [463] validation_0-rmse:0.00682
## [464] validation_0-rmse:0.006792
## [465] validation_0-rmse:0.006725
## [466] validation_0-rmse:0.006659
## [467] validation_0-rmse:0.006593
## [468] validation_0-rmse:0.006567
## [469] validation_0-rmse:0.006502
## [470] validation_0-rmse:0.006438
## [471] validation_0-rmse:0.006374
## [472] validation_0-rmse:0.006311
## [473] validation_0-rmse:0.006249
## [474] validation_0-rmse:0.006187
## [475] validation_0-rmse:0.006162
## [476] validation_0-rmse:0.006102
## [477] validation_0-rmse:0.006078
## [478] validation_0-rmse:0.00606
## [479] validation_0-rmse:0.006042
## [480] validation_0-rmse:0.005983
## [481] validation_0-rmse:0.005961
## [482] validation_0-rmse:0.005902
## [483] validation_0-rmse:0.005843
## [484] validation_0-rmse:0.005786
## [485] validation_0-rmse:0.005729
## [486] validation_0-rmse:0.005672
## [487] validation_0-rmse:0.005616
## [488] validation_0-rmse:0.005561
## [489] validation_0-rmse:0.00554
## [490] validation_0-rmse:0.005486
## [491] validation_0-rmse:0.005431
## [492] validation_0-rmse:0.005417
## [493] validation_0-rmse:0.005363
## [494] validation_0-rmse:0.00531
## [495] validation_0-rmse:0.005258
## [496] validation_0-rmse:0.005239
## [497] validation_0-rmse:0.005187
## [498] validation_0-rmse:0.005136
## [499] validation_0-rmse:0.005085
## [500] validation_0-rmse:0.005068
## [501] validation_0-rmse:0.005018
## [502] validation_0-rmse:0.005001
## [503] validation_0-rmse:0.004952
## [504] validation_0-rmse:0.004903
## [505] validation_0-rmse:0.004886
## [506] validation_0-rmse:0.004838
## [507] validation_0-rmse:0.00479
## [508] validation_0-rmse:0.004743
## [509] validation_0-rmse:0.004696
## [510] validation_0-rmse:0.004681
## [511] validation_0-rmse:0.004635
## [512] validation_0-rmse:0.004589
## [513] validation_0-rmse:0.004574
## [514] validation_0-rmse:0.004529
## [515] validation_0-rmse:0.004515
## [516] validation_0-rmse:0.00447
## [517] validation_0-rmse:0.004426
## [518] validation_0-rmse:0.004383
## [519] validation_0-rmse:0.00434
## [520] validation_0-rmse:0.004297
## [521] validation_0-rmse:0.004254
## [522] validation_0-rmse:0.004241
## [523] validation_0-rmse:0.0042
## [524] validation_0-rmse:0.004158
## [525] validation_0-rmse:0.004117
## [526] validation_0-rmse:0.004076
## [527] validation_0-rmse:0.004064
## [528] validation_0-rmse:0.004024
## [529] validation_0-rmse:0.003985
## [530] validation_0-rmse:0.003976
## [531] validation_0-rmse:0.003937
## [532] validation_0-rmse:0.003898
## [533] validation_0-rmse:0.00386
## [534] validation_0-rmse:0.003822
## [535] validation_0-rmse:0.003784
## [536] validation_0-rmse:0.003747
## [537] validation_0-rmse:0.003736
## [538] validation_0-rmse:0.003699
## [539] validation_0-rmse:0.003663
## [540] validation_0-rmse:0.003627
## [541] validation_0-rmse:0.003591
## [542] validation_0-rmse:0.003556
## [543] validation_0-rmse:0.003521
## [544] validation_0-rmse:0.003486
## [545] validation_0-rmse:0.003476
## [546] validation_0-rmse:0.003442
## [547] validation_0-rmse:0.003432
## [548] validation_0-rmse:0.003423
## [549] validation_0-rmse:0.003414
## [550] validation_0-rmse:0.00338
## [551] validation_0-rmse:0.003347
## [552] validation_0-rmse:0.003338
## [553] validation_0-rmse:0.003305
## [554] validation_0-rmse:0.003296
## [555] validation_0-rmse:0.003288
## [556] validation_0-rmse:0.003255
## [557] validation_0-rmse:0.003223
## [558] validation_0-rmse:0.003192
## [559] validation_0-rmse:0.003184
## [560] validation_0-rmse:0.003176
## [561] validation_0-rmse:0.003145
## [562] validation_0-rmse:0.003113
## [563] validation_0-rmse:0.003083
## [564] validation_0-rmse:0.003052
## [565] validation_0-rmse:0.003045
## [566] validation_0-rmse:0.003015
## [567] validation_0-rmse:0.002986
## [568] validation_0-rmse:0.002979
## [569] validation_0-rmse:0.002949
## [570] validation_0-rmse:0.002943
## [571] validation_0-rmse:0.002936
## [572] validation_0-rmse:0.002907
## [573] validation_0-rmse:0.002879
## [574] validation_0-rmse:0.00285
## [575] validation_0-rmse:0.002844
## [576] validation_0-rmse:0.002838
## [577] validation_0-rmse:0.00281
## [578] validation_0-rmse:0.002782
## [579] validation_0-rmse:0.002755
## [580] validation_0-rmse:0.002728
## [581] validation_0-rmse:0.002701
## [582] validation_0-rmse:0.002675
## [583] validation_0-rmse:0.002672
## [584] validation_0-rmse:0.002646
## [585] validation_0-rmse:0.00262
## [586] validation_0-rmse:0.002594
## [587] validation_0-rmse:0.002569
## [588] validation_0-rmse:0.002543
## [589] validation_0-rmse:0.002538
## [590] validation_0-rmse:0.002513
## [591] validation_0-rmse:0.002488
## [592] validation_0-rmse:0.002483
## [593] validation_0-rmse:0.002459
## [594] validation_0-rmse:0.002435
## [595] validation_0-rmse:0.002411
## [596] validation_0-rmse:0.002387
## [597] validation_0-rmse:0.002364
## [598] validation_0-rmse:0.002362
## [599] validation_0-rmse:0.002339
## [600] validation_0-rmse:0.002316
## [601] validation_0-rmse:0.002293
## [602] validation_0-rmse:0.00227
## [603] validation_0-rmse:0.002266
## [604] validation_0-rmse:0.002244
## [605] validation_0-rmse:0.002222
## [606] validation_0-rmse:0.0022
## [607] validation_0-rmse:0.002178
## [608] validation_0-rmse:0.002174
## [609] validation_0-rmse:0.002153
## [610] validation_0-rmse:0.002132
## [611] validation_0-rmse:0.002128
## [612] validation_0-rmse:0.002107
## [613] validation_0-rmse:0.002086
## [614] validation_0-rmse:0.002083
## [615] validation_0-rmse:0.002062
## [616] validation_0-rmse:0.002042
## [617] validation_0-rmse:0.002022
## [618] validation_0-rmse:0.002018
## [619] validation_0-rmse:0.001999
## [620] validation_0-rmse:0.001995
## [621] validation_0-rmse:0.001976
## [622] validation_0-rmse:0.001956
## [623] validation_0-rmse:0.001953
## [624] validation_0-rmse:0.001934
## [625] validation_0-rmse:0.001915
## [626] validation_0-rmse:0.001896
## [627] validation_0-rmse:0.001877
## [628] validation_0-rmse:0.001859
## [629] validation_0-rmse:0.001841
## [630] validation_0-rmse:0.001823
## [631] validation_0-rmse:0.001805
## [632] validation_0-rmse:0.001787
## [633] validation_0-rmse:0.00177
## [634] validation_0-rmse:0.001752
## [635] validation_0-rmse:0.001735
## [636] validation_0-rmse:0.001718
## [637] validation_0-rmse:0.001701
## [638] validation_0-rmse:0.001685
## [639] validation_0-rmse:0.001684
## [640] validation_0-rmse:0.001668
## [641] validation_0-rmse:0.001651
## [642] validation_0-rmse:0.001635
## [643] validation_0-rmse:0.001619
## [644] validation_0-rmse:0.001603
## [645] validation_0-rmse:0.001601
## [646] validation_0-rmse:0.001598
## [647] validation_0-rmse:0.001583
## [648] validation_0-rmse:0.001567
## [649] validation_0-rmse:0.001552
## [650] validation_0-rmse:0.001537
## [651] validation_0-rmse:0.001522
## [652] validation_0-rmse:0.001507
## [653] validation_0-rmse:0.001492
## [654] validation_0-rmse:0.001478
## [655] validation_0-rmse:0.001463
## [656] validation_0-rmse:0.001449
## [657] validation_0-rmse:0.001435
## [658] validation_0-rmse:0.001421
## [659] validation_0-rmse:0.001418
## [660] validation_0-rmse:0.001405
## [661] validation_0-rmse:0.001391
## [662] validation_0-rmse:0.001377
## [663] validation_0-rmse:0.001364
## [664] validation_0-rmse:0.001351
## [665] validation_0-rmse:0.001337
## [666] validation_0-rmse:0.001324
## [667] validation_0-rmse:0.001312
## [668] validation_0-rmse:0.001299
## [669] validation_0-rmse:0.001286
## [670] validation_0-rmse:0.001274
## [671] validation_0-rmse:0.001261
## [672] validation_0-rmse:0.001249
## [673] validation_0-rmse:0.001237
## [674] validation_0-rmse:0.001225
## [675] validation_0-rmse:0.001213
## [676] validation_0-rmse:0.001211
## [677] validation_0-rmse:0.001209
## [678] validation_0-rmse:0.001207
## [679] validation_0-rmse:0.001206
## [680] validation_0-rmse:0.001194
## [681] validation_0-rmse:0.001192
## [682] validation_0-rmse:0.001191
## [683] validation_0-rmse:0.001179
## [684] validation_0-rmse:0.001177
## [685] validation_0-rmse:0.001178
## [686] validation_0-rmse:0.001166
## [687] validation_0-rmse:0.001155
## [688] validation_0-rmse:0.001153
## [689] validation_0-rmse:0.001152
## [690] validation_0-rmse:0.001151
## [691] validation_0-rmse:0.001139
## [692] validation_0-rmse:0.001128
## [693] validation_0-rmse:0.001117
## [694] validation_0-rmse:0.001106
## [695] validation_0-rmse:0.001096
## [696] validation_0-rmse:0.001085
## [697] validation_0-rmse:0.001075
## [698] validation_0-rmse:0.001064
## [699] validation_0-rmse:0.001063
## [700] validation_0-rmse:0.001053
## [701] validation_0-rmse:0.001042
## [702] validation_0-rmse:0.001032
## [703] validation_0-rmse:0.001031
## [704] validation_0-rmse:0.001032
## [705] validation_0-rmse:0.001022
## [706] validation_0-rmse:0.001012
## [707] validation_0-rmse:0.001011
## [708] validation_0-rmse:0.001001
## [709] validation_0-rmse:0.001
## [710] validation_0-rmse:0.00099
## [711] validation_0-rmse:0.00098
## [712] validation_0-rmse:0.000971
## [713] validation_0-rmse:0.00097
## [714] validation_0-rmse:0.000969
## [715] validation_0-rmse:0.000968
## [716] validation_0-rmse:0.000959
## [717] validation_0-rmse:0.000949
## [718] validation_0-rmse:0.000948
## [719] validation_0-rmse:0.000949
## [720] validation_0-rmse:0.00094
## [721] validation_0-rmse:0.000931
## [722] validation_0-rmse:0.000922
## [723] validation_0-rmse:0.000913
## [724] validation_0-rmse:0.000912
## [725] validation_0-rmse:0.000903
## [726] validation_0-rmse:0.000895
## [727] validation_0-rmse:0.000886
## [728] validation_0-rmse:0.000877
## [729] validation_0-rmse:0.000869
## [730] validation_0-rmse:0.00086
## [731] validation_0-rmse:0.00086
## [732] validation_0-rmse:0.000851
## [733] validation_0-rmse:0.000843
## [734] validation_0-rmse:0.000835
## [735] validation_0-rmse:0.000827
## [736] validation_0-rmse:0.000828
## [737] validation_0-rmse:0.00082
## [738] validation_0-rmse:0.000819
## [739] validation_0-rmse:0.000811
## [740] validation_0-rmse:0.000803
## [741] validation_0-rmse:0.000796
## [742] validation_0-rmse:0.000788
## [743] validation_0-rmse:0.00078
## [744] validation_0-rmse:0.000773
## [745] validation_0-rmse:0.000774
## [746] validation_0-rmse:0.000773
## [747] validation_0-rmse:0.000766
## [748] validation_0-rmse:0.000758
## [749] validation_0-rmse:0.000751
## [750] validation_0-rmse:0.000744
## [751] validation_0-rmse:0.000737
## [752] validation_0-rmse:0.00073
## [753] validation_0-rmse:0.000723
## [754] validation_0-rmse:0.000723
## [755] validation_0-rmse:0.000717
## [756] validation_0-rmse:0.000716
## [757] validation_0-rmse:0.000709
## [758] validation_0-rmse:0.000702
## [759] validation_0-rmse:0.000696
## [760] validation_0-rmse:0.000689
## [761] validation_0-rmse:0.000682
## [762] validation_0-rmse:0.000683
## [763] validation_0-rmse:0.000683
## [764] validation_0-rmse:0.000683
## [765] validation_0-rmse:0.000677
## [766] validation_0-rmse:0.00067
## [767] validation_0-rmse:0.000664
## [768] validation_0-rmse:0.000658
## [769] validation_0-rmse:0.000659
## [770] validation_0-rmse:0.000652
## [771] validation_0-rmse:0.000646
## [772] validation_0-rmse:0.00064
## [773] validation_0-rmse:0.000634
## [774] validation_0-rmse:0.000633
## [775] validation_0-rmse:0.000627
## [776] validation_0-rmse:0.000627
## [777] validation_0-rmse:0.000621
## [778] validation_0-rmse:0.00062
## [779] validation_0-rmse:0.000614
## [780] validation_0-rmse:0.000609
## [781] validation_0-rmse:0.000603
## [782] validation_0-rmse:0.000604
## [783] validation_0-rmse:0.000603
## [784] validation_0-rmse:0.000603
## [785] validation_0-rmse:0.000597
## [786] validation_0-rmse:0.000591
## [787] validation_0-rmse:0.000591
## [788] validation_0-rmse:0.000585
## [789] validation_0-rmse:0.00058
## [790] validation_0-rmse:0.000574
## [791] validation_0-rmse:0.000574
## [792] validation_0-rmse:0.000568
## [793] validation_0-rmse:0.000563
## [794] validation_0-rmse:0.000558
## [795] validation_0-rmse:0.000552
## [796] validation_0-rmse:0.000547
## [797] validation_0-rmse:0.000542
## [798] validation_0-rmse:0.000537
## [799] validation_0-rmse:0.000532
## [800] validation_0-rmse:0.000527
## [801] validation_0-rmse:0.000526
## [802] validation_0-rmse:0.000526
## [803] validation_0-rmse:0.000521
## [804] validation_0-rmse:0.000521
## [805] validation_0-rmse:0.000516
## [806] validation_0-rmse:0.000515
## [807] validation_0-rmse:0.000511
## [808] validation_0-rmse:0.000506
## [809] validation_0-rmse:0.000501
## [810] validation_0-rmse:0.000496
## [811] validation_0-rmse:0.000492
## [812] validation_0-rmse:0.000492
## [813] validation_0-rmse:0.000488
## [814] validation_0-rmse:0.000483
## [815] validation_0-rmse:0.000483
## [816] validation_0-rmse:0.000484
## [817] validation_0-rmse:0.000479
## [818] validation_0-rmse:0.000474
## [819] validation_0-rmse:0.00047
## [820] validation_0-rmse:0.000471
## [821] validation_0-rmse:0.000471
## [822] validation_0-rmse:0.000467
## [823] validation_0-rmse:0.000463
## [824] validation_0-rmse:0.000458
## [825] validation_0-rmse:0.000454
## [826] validation_0-rmse:0.00045
## [827] validation_0-rmse:0.000446
## [828] validation_0-rmse:0.000441
## [829] validation_0-rmse:0.000441
## [830] validation_0-rmse:0.000437
## [831] validation_0-rmse:0.000433
## [832] validation_0-rmse:0.000429
## [833] validation_0-rmse:0.000425
## [834] validation_0-rmse:0.000424
## [835] validation_0-rmse:0.00042
## [836] validation_0-rmse:0.000417
## [837] validation_0-rmse:0.000413
## [838] validation_0-rmse:0.000409
## [839] validation_0-rmse:0.000408
## [840] validation_0-rmse:0.000405
## [841] validation_0-rmse:0.000404
## [842] validation_0-rmse:0.000404
## [843] validation_0-rmse:0.0004
## [844] validation_0-rmse:0.0004
## [845] validation_0-rmse:0.000396
## [846] validation_0-rmse:0.000393
## [847] validation_0-rmse:0.000389
## [848] validation_0-rmse:0.00039
## [849] validation_0-rmse:0.000386
## [850] validation_0-rmse:0.000386
## [851] validation_0-rmse:0.000382
## [852] validation_0-rmse:0.000379
## [853] validation_0-rmse:0.000379
## [854] validation_0-rmse:0.000375
## [855] validation_0-rmse:0.000375
## [856] validation_0-rmse:0.000371
## [857] validation_0-rmse:0.000368
## [858] validation_0-rmse:0.000368
## [859] validation_0-rmse:0.000364
## [860] validation_0-rmse:0.000365
## [861] validation_0-rmse:0.000365
## [862] validation_0-rmse:0.000361
## [863] validation_0-rmse:0.000358
## [864] validation_0-rmse:0.000355
## [865] validation_0-rmse:0.000351
## [866] validation_0-rmse:0.000351
## [867] validation_0-rmse:0.000348
## [868] validation_0-rmse:0.000345
## [869] validation_0-rmse:0.000345
## [870] validation_0-rmse:0.000342
## [871] validation_0-rmse:0.000339
## [872] validation_0-rmse:0.000336
## [873] validation_0-rmse:0.000334
## [874] validation_0-rmse:0.000333
## [875] validation_0-rmse:0.000331
## [876] validation_0-rmse:0.00033
## [877] validation_0-rmse:0.000329
## [878] validation_0-rmse:0.000327
## [879] validation_0-rmse:0.000326
## [880] validation_0-rmse:0.000324
## [881] validation_0-rmse:0.000323
## [882] validation_0-rmse:0.000322
## [883] validation_0-rmse:0.000321
## [884] validation_0-rmse:0.000319
## [885] validation_0-rmse:0.000318
## [886] validation_0-rmse:0.000317
## [887] validation_0-rmse:0.000316
## [888] validation_0-rmse:0.000315
## [889] validation_0-rmse:0.000313
## [890] validation_0-rmse:0.000312
## [891] validation_0-rmse:0.000311
## [892] validation_0-rmse:0.00031
## [893] validation_0-rmse:0.000309
## [894] validation_0-rmse:0.000308
## [895] validation_0-rmse:0.000307
## [896] validation_0-rmse:0.000306
## [897] validation_0-rmse:0.000305
## [898] validation_0-rmse:0.000304
## [899] validation_0-rmse:0.000303
## [900] validation_0-rmse:0.000303
## [901] validation_0-rmse:0.000302
## [902] validation_0-rmse:0.000302
## [903] validation_0-rmse:0.000301
## [904] validation_0-rmse:0.0003
## [905] validation_0-rmse:0.000299
## [906] validation_0-rmse:0.000299
## [907] validation_0-rmse:0.000298
## [908] validation_0-rmse:0.000298
## [909] validation_0-rmse:0.000298
## [910] validation_0-rmse:0.000297
## [911] validation_0-rmse:0.000297
## [912] validation_0-rmse:0.000296
## [913] validation_0-rmse:0.000295
## [914] validation_0-rmse:0.000294
## [915] validation_0-rmse:0.000295
## [916] validation_0-rmse:0.000294
## [917] validation_0-rmse:0.000293
## [918] validation_0-rmse:0.000293
## [919] validation_0-rmse:0.000293
## [920] validation_0-rmse:0.000293
## [921] validation_0-rmse:0.000293
## [922] validation_0-rmse:0.000292
## [923] validation_0-rmse:0.000292
## [924] validation_0-rmse:0.000292
## [925] validation_0-rmse:0.000292
## [926] validation_0-rmse:0.000291
## [927] validation_0-rmse:0.000291
## [928] validation_0-rmse:0.00029
## [929] validation_0-rmse:0.000289
## [930] validation_0-rmse:0.000289
## [931] validation_0-rmse:0.000288
## [932] validation_0-rmse:0.000287
## [933] validation_0-rmse:0.000287
## [934] validation_0-rmse:0.000287
## [935] validation_0-rmse:0.000286
## [936] validation_0-rmse:0.000286
## [937] validation_0-rmse:0.000285
## [938] validation_0-rmse:0.000286
## [939] validation_0-rmse:0.000285
## [940] validation_0-rmse:0.000285
## [941] validation_0-rmse:0.000286
## [942] validation_0-rmse:0.000286
## [943] validation_0-rmse:0.000285
## [944] validation_0-rmse:0.000285
## [945] validation_0-rmse:0.000285
## [946] validation_0-rmse:0.000284
## [947] validation_0-rmse:0.000284
## [948] validation_0-rmse:0.000283
## [949] validation_0-rmse:0.000282
## [950] validation_0-rmse:0.000282
## [951] validation_0-rmse:0.000282
## [952] validation_0-rmse:0.000281
## [953] validation_0-rmse:0.000281
## [954] validation_0-rmse:0.00028
## [955] validation_0-rmse:0.00028
## [956] validation_0-rmse:0.000279
## [957] validation_0-rmse:0.000279
## [958] validation_0-rmse:0.000279
## [959] validation_0-rmse:0.000278
## [960] validation_0-rmse:0.000278
## [961] validation_0-rmse:0.000277
## [962] validation_0-rmse:0.000277
## [963] validation_0-rmse:0.000277
## [964] validation_0-rmse:0.000277
## [965] validation_0-rmse:0.000277
## [966] validation_0-rmse:0.000278
## [967] validation_0-rmse:0.000277
## [968] validation_0-rmse:0.000277
## [969] validation_0-rmse:0.000277
## [970] validation_0-rmse:0.000277
## [971] validation_0-rmse:0.000277
## Stopping. Best iteration:
## [961] validation_0-rmse:0.000277
##
## XGBRegressor(base_score=0.5, booster='gbtree', colsample_bylevel=1,
## colsample_bynode=1, colsample_bytree=0.8, gamma=0,
## importance_type='gain', learning_rate=0.01, max_delta_step=0,
## max_depth=10, min_child_weight=9, missing=None, n_estimators=10000,
## n_jobs=1, nthread=8, objective='reg:logistic', random_state=0,
## reg_alpha=0, reg_lambda=1, scale_pos_weight=1, seed=4242,
## silent=None, subsample=0.8, verbosity=1)## 4.1780074 Minutes# Predict
preds = clf.predict(val_x)
## Wich max prob
import numpy as np
best_preds = np.asarray([np.argmax(line) for line in preds])
# Determine the precision of this prediction:
from sklearn.metrics import precision_score
precision_score(val_y, best_preds, average='macro')## 0.46495489863356254
##
## /Users/Mezhoud/anaconda3/lib/python3.7/site-packages/sklearn/metrics/_classification.py:1272: UndefinedMetricWarning: Precision is ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.
## _warn_prf(average, modifier, msg_start, len(result))#xtest = test.drop(['Target', 'weeks', 'Latitude', 'Longitude'], axis=1)
test_ID = xtest.Square_ID
xtest = xtest.drop('Square_ID', axis = 1)
pred_test = clf.predict(xtest)submission = pd.concat([pd.DataFrame(test_ID), pd.DataFrame(pred_test)], axis = 1)
submission.columns = ['Square_ID', 'pred']
submission.to_csv("pred_test.csv")
submission.head## <bound method NDFrame.head of Square_ID pred
## 0 4e3c3896 0.000041
## 1 4e3c3897 0.000041
## 2 4e3c3898 0.000041
## 3 4e3c3899 0.000041
## 4 4e3c389a 0.000041
## ... ... ...
## 279917 4e6f5dfd 0.000042
## 279918 4e6f5dfe 0.000042
## 279919 4e6f5dff 0.000042
## 279920 4e6f5e00 0.000042
## 279921 4e6f5e01 0.000047
##
## [279922 rows x 2 columns]>7.1 Visualize prediction
pred_test <- fread("pred_test.csv")
new_train[new_train$Year == 2019,]$Target <- pred_test$pred
ggplot(new_train, aes(Target)) + geom_density()
new_train %>%
#filter(Year == 2019) %>%
group_by(Year) %>%
filter(`Week of` == max(`Week of`))%>%
ungroup()%>%
mutate(Height = as.factor(Height)) %>%
ggplot +
#geom_point(aes(x = Longitude, y = Latitude, colour = Height)) +
geom_point(#data = subset(new_train, Target == 1),
aes(x = Longitude, y = Latitude, color = Target, shape = Target),
size = 0.8, stroke = 0, shape = 16) +
facet_wrap(Year~., ncol = 2)